Guide
Seedance 2.5 Flash Text-to-Video: Parameters, Per-Second Billing and Production Practice
From the first request to a stable pipeline: how to set parameters and how the bill adds up
- Modell
- seedance-2.5-flash
Auf dieser Seite
What this model does
seedance-2.5-flash/text-to-video is ByteDance's text-to-video model: you send a text prompt and get back a video with synchronized audio. That is its only input path — it does not take images or video as conditioning, and it does not do video understanding.
It fits "one sentence in, one usable short clip out": batch marketing assets, atmospheric inserts for product films, filling a content pool with short pieces, and quick creative previews. If your starting point is an image, the image-to-video route is more direct; if you need a subject to stay consistent with supplied reference material, use the reference-driven route.
The request parameters
- prompt (required): scene, subject, camera movement and action. Up to 5,000 characters.
- resolution:
480por720p, default480p. - duration: an integer from 4 to 30, default 5, in seconds.
- aspect_ratio:
16:9,4:3,1:1,3:4,9:16,21:9,adaptive; defaultadaptive, which lets the model pick a ratio that fits the prompt. - output_format:
mp4(default) ormov. - generate_audio: defaults to
trueand produces synchronized dialogue, sound effects and background audio.
Everything except prompt has a default, so the smallest request is just a prompt. The request body is validated strictly: undeclared fields are rejected rather than silently dropped, so integration mistakes surface at submission time instead of after you have been billed.
How per-second billing works
You pay the unit price for the resolution tier multiplied by the number of output seconds:
- 480p: $0.0679/second
- 720p: $0.1525/second
Output seconds are the duration you request. A 5-second 480p clip is $0.3395 and a 5-second 720p clip is $0.7625; 30 seconds at 480p is $2.037 and at 720p is $4.575. Double the duration and you double the cost — there is no volume discount for longer clips.
Three practical consequences: budgeting is multiplication, so fix the per-clip duration first and multiply by clip count; a retry is a second charge, so cap retries; and long pieces should be generated as several short clips and edited together, which costs less and limits a failure to the one bad segment.
Integration shape: asynchronous tasks
Video generation takes time, so the API is asynchronous: submit, receive a task identifier, then poll or wait for a callback, and finally download and store the result. Do not wrap it as a synchronous RPC that blocks the user's request.
- Persist the state machine. Task id, status and request parameters belong in a database, not process memory — a restart must not lose a task someone already paid for.
- Back off on polling. Start with short intervals, lengthen them with jitter, and give every task a total timeout that lands in a failure branch instead of polling forever.
- Classify errors. Invalid parameters and content-policy rejections are deterministic failures; retrying them only burns budget. Network blips and rate limits are the ones worth retrying with backoff.
- Be idempotent. Send a business unique key on submit so a double click does not create two tasks and two charges.
Writing prompts that hold up
Write four blocks — subject, action sequence, environment and light, camera — instead of stacking adjectives.
Subject: a white cat sitting on a windowsill, backlit
Action: from stillness to lifting its head toward the window, tail swaying slowly
Environment and light: warm afternoon light, shallow depth of field, quiet indoor mood
Camera: medium shot, locked-off, slight handheld feel
Three rules: one prompt, one continuous shot — do not pack scene changes into a single generation; version your prompts and change one variable at a time so you can tell which edit mattered; validate on a small sample before scaling, never use volume to test an undecided prompt.
Describe framing and camera language in natural language. There is no separate camera parameter, so putting it in the prompt is more reliable than hoping for a field that does not exist.
Path to production
- Prove the pipeline. Three prompts covering typical scenes, confirming submit, poll and download all work. Do not judge quality yet.
- Fix prompts and duration. Hold a small set of prompts constant, compare writes and durations, log the cost of each, and build a reusable template library.
- Ramp gradually. Shift traffic in stages and watch success rate, latency, retry rate and the cost curve; add volume only when all three are on target.
- Operate it. Review the template library, retire weak prompts, and keep an eye on cost trend and retry thresholds.
The model page carries the current parameter reference and prices — check it once before you integrate.
Ähnliche Artikel
Alle Artikel ansehen →-
Guide
Seedance 2.5 Flash Reference-to-Video: Reference Images, Audio and How the Bill Is Counted
9 min read -
Guide
Seedance 2.5 Flash Image-to-Video: First Frame, Last Frame and Per-Second Billing
Making a still image move: parameters, price and the usual failure modes
7 min read -
Guide
kling-image-4.0 Getting Started Guide: Key Integration Points for the Image Generation API and Cost Control
Integration decisions and practical checklists for developers